BlackRock’s Aladdin platform ran scenario forecasts on roughly twenty-one trillion dollars of assets across one Tuesday in May 2025, returning value-at-risk and stress numbers before US markets opened. That figure, larger than the GDP of any country except the US itself, is the working stake for predictive modeling america in 2026. The biggest US banks and asset managers now treat predictive models the way they once treated phone lines, as the connective tissue of every decision. This piece walks through where those models pay off, where they hurt, and what comes next.
Use cases the largest US firms already depend on
Consumer credit is the first. Capital One built a fifteen-year head start on machine learning credit models, and the bank now uses gradient-boosted models for credit card, auto, and small business underwriting. Each model is trained on tens of millions of accounts and refreshed monthly. JPMorgan, Discover, and Synchrony run comparable stacks. The output drives roughly one hundred and seventy billion dollars of US credit card balances added each year, with loss rates below the levels US bank supervisors expected coming out of the pandemic.
Payment fraud is the second. Stripe Radar, the in-house fraud product at Stripe, scores every authorization the company processes for more than four million US businesses. Radar uses an ensemble of models trained on the network’s full transaction history, which now runs into the high hundreds of billions of dollars per year. The result, according to Federal Reserve payment data on the Federal Reserve payments page, is that US card fraud loss rates have held near eight basis points of volume even as authorization counts climbed past one hundred fifty billion per year.
Asset management is the third. BlackRock’s Aladdin runs scenario forecasts and risk decompositions on a meaningful share of US institutional portfolios. The platform is licensed to peer asset managers, US pension funds, and US insurers, and it is the single best-known example of a US firm selling predictive modelling as infrastructure. State Street’s Charles River and SS&C’s Eze do similar work for other parts of the market. The combined coverage runs into the tens of trillions of dollars under model.
Bank treasury is the fourth. Every large US bank runs deposit beta models that predict how deposit balances respond to Federal Reserve rate moves. After the March 2023 regional bank failures, supervisors pushed for better intraday liquidity modelling, and the largest US banks now refresh deposit and liquidity forecasts continuously. FDIC industry data on the quarterly banking profile tracks the deposit movement that those models try to predict, and the TechBullion regtech compliance overview covers how the largest US banks fold these signals into supervisory reporting.
The benefits US firms count in dollars and basis points
The first benefit is loss reduction. Capital One has been public that machine learning credit models reduced expected losses on its card book by hundreds of basis points relative to the traditional scorecard approach. Synchrony and Discover have made similar claims at investor days. Compounded across an industry that issues nearly four trillion dollars of US consumer credit, the savings are large.
The second benefit is faster product launches. A US fintech with a working data platform and a trained risk modeller can launch a new credit product in months rather than years. Affirm, Klarna’s US arm, and the buy-now-pay-later cohort built their books on this speed advantage. McKinsey’s financial services research on the McKinsey financial services insights page has put the time-to-market advantage at four to ten times for firms that have completed a full model deployment stack.
The third benefit is portfolio sensitivity. Asset managers that run scenario forecasts continuously can respond to macro shifts in hours rather than weeks. The 2023 regional bank stress was a live test of this capability, and US managers that had wired their portfolios into modern scenario systems repositioned faster than those that had not. The TechBullion coverage in the fintech news hub tracks the US firms that have built this capacity.
Risks the largest US firms cannot ignore
The first risk is model drift. The 2022 inflation spike, the 2023 regional bank stress, and the 2024 synthetic identity wave each surprised US predictive models that had not been retrained. SR 11-7, the Federal Reserve and OCC model risk guidance, expects banks to monitor drift, but vendor models often arrive with limited visibility into training data, which makes drift monitoring harder than the guidance presumes.
The second risk is concentration. Aladdin’s footprint inside US asset management has drawn regulatory attention because so many US institutional portfolios are scored on a single platform. If the platform’s underlying assumptions move in one direction, a meaningful share of US asset managers may move with it. The Office of Financial Research at Treasury has flagged platform concentration in published work.
The third risk is fair lending. The Consumer Financial Protection Bureau has been clear, through circulars and supervisory actions, that US lenders own the fair lending risk of every predictive model they deploy. The CFPB’s research reports on the CFPB research reports page have documented disparate outcomes in auto, small business, and consumer credit that traced back to model inputs and to the data the model was trained on. The 2022 CFPB circular on adverse action notices made plain that a US lender has to disclose specific denial reasons even when the underlying model is a black box.
The fourth risk is talent. The Bureau of Labor Statistics tracks growth in software and data roles in its published occupational outlook, with median pay above one hundred thirty thousand dollars and growth above twenty percent through 2032. US banks pay below the largest US technology firms for senior modellers, which means trained staff often leave after eighteen months. The departure rate slows every other initiative on this list.
Long-term opportunities for US predictive modelling
The first opportunity is climate and weather data. US property and casualty insurers have already wired weather feeds into pricing. The largest US banks, with mortgage and commercial real estate books exposed to flood, fire, and hurricane risk, are following. Several large US lenders have published climate risk integration plans aligned with Federal Reserve scenario analysis. Predictive models that combine traditional credit inputs with climate exposures are likely to differentiate underwriters over the next decade.
The second opportunity is small business cash flow lending. With accounting and bank data flowing through APIs, US fintechs and community banks can build predictive models that score a small business in real time on the last ninety days of activity rather than on a bureau pull and tax returns. Pilots at several US community banks have shown loss rates within striking distance of larger banks despite faster approvals.
The third opportunity is intraday risk for US clearing and prime brokerage. The volume of margin calls in the 2024 first quarter, during a volatile rates and equity period, was a stress test for the current generation of risk models at the largest US dealers. The next generation, trained on streaming data with continuous scoring, will let US clearing firms reduce procyclical margin calls. The TechBullion payments hub covers the operational shift across the US clearing stack.
What to watch over the next twenty-four months
Three signals will set the pace. First, the CFPB’s open banking rule, when final, expands cash flow data into mainstream US credit underwriting, which is the single largest external lever on predictive modelling adoption in the next cycle. Second, the National Institute of Standards and Technology AI Risk Management Framework adoption pattern at the largest US banks will shape the documentation and validation expectations under which every US bank model has to operate. NIST AI RMF is voluntary, but Federal Reserve and OCC supervisory letters in 2025 have begun citing it as the benchmark. Third, Federal Reserve and OCC commentary on third-party concentration, including platform concentration at firms like BlackRock’s Aladdin and Stripe Radar, will set the boundaries on how much of US finance can run on a single vendor’s models. The next four quarters of supervisory letters, CFPB rule activity, and NIST adoption will decide whether US predictive modelling becomes the foundation of a faster credit and capital markets system, or a regulated specialty under heavier reporting requirements.
